Pete Kowalczyk – Conversation Designer, GOV.UK AI team, GDS

Creating tools for scaling generated content

Posted by: and , Posted on: - Categories: Content Design, Design, Experimental
A flow diagram. The plan is then checked by a rule engine with 21 criteria and a model acting as judge with 19 criteria. This gives a score for each plan, such as 54 out of 56. One loop lets you change the prompt and regenerate a single plan. The other reads all the failures for a rubric and suggests one change to the prompt.

The GOV.UK AI team shares their experimental work on curating content for users to create personalised experiences. They share how they've been using tools to measure the quality of scaled generated content.